By Habib Ur Rehman
Publication Date: 2026-08-27 17:58:00
OpenAI says its first homegrown inference chip can do up to 1.9 times more AI work per watt than the comparison systems, and it went from initial design to tape-out in nine months. The more investable surprise is who helped turn that sketch into silicon.
Broadcom Inc. (NASDAQ:AVGO) handled Jalapeño’s physical implementation and supplied the Tomahawk networking. Celestica is responsible for board, rack and system integration, while TSMC manufactured the chip. OpenAI designed the architecture and said its own models helped it move from initial design to tape-out in nine months, much faster than usual.
OpenAI’s Nvidia Win Is Narrower Than It Sounds
In OpenAI’s published InferenceX results, Jalapeño delivered 1.5 to 1.9 times more work per watt and 1.7 to 3.6 times lower end-to-end latency across three large open models. The comparisons included Nvidia GB200 and GB300 systems.
